A Cloud Detection Network Based on Adaptive Laplacian Coordination Enhanced Cross-Feature U-Net

Area of Science:

• Atmospheric Science
• Renewable Energy Technology
• Computer Vision

Background:

• Cloud cover variability significantly impacts solar irradiance and photovoltaic (PV) power output.
• Precise detection of thin, fragmented clouds is essential for reliable PV power forecasting.

Purpose of the Study:

• To introduce a novel cloud detection method, ALCU-Net, for improved accuracy in PV power generation forecasting.
• To enhance the U-Net architecture with specialized modules for better cloud feature extraction and spatial coherence.

Main Methods:

• Developed Adaptive Laplacian Coordination Enhanced Cross-Feature U-Net (ALCU-Net).
• Incorporated Adaptive Feature Coordination (AFC), Multi-Grained Laplacian-Enhanced (MLE) features, and Criss-Cross Feature Fused Detection (CCFE) modules.
• Augmented traditional U-Net with enhanced spatial coherence, hierarchical feature integration, and refined edge detection.

Main Results:

• ALCU-Net demonstrated superior performance compared to existing cloud detection methods.
• Achieved high accuracy in identifying both thick and thin clouds.
• Successfully mapped fragmented cloud patches across diverse environments (ocean, polar, ocean-land mixtures).

Conclusions:

• ALCU-Net offers a significant advancement in cloud detection for solar energy applications.
• The method's robustness across various environments makes it suitable for real-world PV forecasting.
• Improved cloud detection accuracy directly translates to more reliable photovoltaic power generation predictions.